LIN28 Selectively Modulates a Subclass of Let-7 MicroRNAs

Abstract

LIN28 is a bipartite RNA-binding protein that post-transcriptionally inhibits the biogenesis of let-7 microRNAs to regulate development and influence disease states. However, the mechanisms of let-7 suppression remain poorly understood because LIN28 recognition depends on coordinated targeting by both the zinc knuckle domain (ZKD), which binds a GGAG-like element in the precursor, and the cold shock domain (CSD), whose binding sites have not been systematically characterized. By leveraging single-nucleotide-resolution mapping of LIN28 binding sites in vivo, we determined that the CSD recognizes a (U)GAU motif. This motif partitions the let-7 microRNAs into two subclasses, precursors with both CSD and ZKD binding sites (CSD⁺) and precursors with ZKD but no CSD binding sites (CSD^-). LIN28 in vivo recognition-and subsequent 3' uridylation and degradation-of CSD⁺ precursors is more efficient, leading to their stronger suppression in LIN28-activated cells and cancers. Thus, CSD binding sites amplify the regulatory effects of LIN28.

Keywords: CLIP; LIN28; bipartite binding; cancer; cold shock domain; let-7 microRNA biogenesis; selective suppression; stem cell.

Figure 1:. LIN28 cold shock domain (CSD) and zinc knuckle domain (ZKD) recognize distinct sequence motifs as defined by single-nucleotide-resolution analysis of CLIP data.

Related to Figures S1. (A) Schematic representation of LIN28 protein domains. (B, C) The ZKD and CSD binding motifs determined from single-nucleotide-resolution analysis of CLIP data. A GGAG-like motif was identified by modeling sequences around LIN28A CIMS derived from mouse ESCs (B, right panel), and a UGAU motif was determined by modeling sequences around LIN28B CITS derived from K562 cells (C, right panel). The frequency of crosslinking at each motif position is shown under the motif logos. The enrichment of GGAG and UGAU tetramers around CIMS or CITS is shown on the left of each panel. (D, E) The crystal structure of LIN28A ZKD (D) and CSD (E) in complex with let-7g hairpin (PDB accession: 3TS2). Residues that are in direct contact with RNA are highlighted in blue. The crosslinked nucleotides are indicated in red and highlighted. (F) Frequency of tetramers conforming to the NGAU consensus in LIN28B eCLIP data from K562 cells. The fold enrichment of each tetramer at the crosslink site in comparison to matched control sequences is shown in the parentheses. (G) CSD binding motifs identified from RBNS analysis. The most enriched pentamers and hexamers after two rounds of LIN28A CSD selection are shown. (H) Enrichment of NGAU and GGAG around LIN28 eCLIP tag cluster peaks from K562 cells.

Figure 2:. The cold shock domain modulates LIN28 binding to CSD⁺ let-7 precursors.

Related to Figure S2. (A) Multiple sequence alignments of pre-let-7 hairpins. Sequences corresponding to mature miRNAs, and binding sites of LIN28 CSD and ZKD are indicated. Let-7 family members are divided into two subclasses, denoted CSD⁺ and CSD⁻, depending on the presence of the GAU (GAC in the case of let-7i) motif. Mutant CSD⁺ let-7 precursors tested in this study are also shown. (B) Quantification of LIN28B binding to let-7 pre-miRNAs in K562 cells. The y-axis shows the total number of unique CLIP tags expressed in reads per million (RPM) that overlap with each pre-let-7. The x-axis shows the number of mock CLIP tags (input) expressed in RPM reflecting the abundance of the pre-let-7. ANOVA was used to test the difference in LIN28 binding to the CSD⁺ versus CSD⁻ pre-let-7s after controlling for pre-miRNA abundance. (C) LIN28 binding to different let-7 family members in human HepG2 and K562 cells using the let-7a-1/7f-1/7d poly-cistronic miRNA locus as an example. The number of mock (gray) and IP (green) tags in each genomic position is shown, and the locations of the pre-miRNA hairpins are indicated at the bottom. (D) RNA-mediated LIN28A/B pull-down using different pri-let-7 family hairpins as a bait quantified by mass spectrometry. The normalized spectrum counts of mass spectrometry-identified peptides from LIN28A (left) or LIN28B (right) are shown for each bait and compared between the CSD⁺ and CSD⁻ subclasses. The boxplots indicate the interquartile range of each subclass. The difference between the two subclasses was evaluated by a t-test. (E) RNA-mediated LIN28A pull-down using different pri-let-7 family hairpins as a bait quantified by immunoblots. LIN28 intensity detected using a specific antibody was normalized using northern blot signal for each individual bait. CSD⁺ hairpins are shown in blue and CSD⁻ hairpins are shown in red, respectively. pri-miR-18b is used as a negative control. Error bars represent standard error of the mean (SEM) of two replicates. Comparison of CSD⁺ and CSD⁻ hairpins was performed using ANOVA of a linear mixed effect model. (F) RNA-mediated LIN28A pull-down using wild type (WT) and mutant (Mut) pri-let-7g and pri-mi-R98 hairpins. The amount of bound LIN28 is quantified as in (E). Reduction of the LIN28 in the mutant is compared to the wild type of the corresponding miRNA precursor using a single-sided t-test. Error bars represent SEM of two replicates.

Figure 3:. Selective 3ʹ polyuridylation and suppression of CSD⁺ let-7 in human cells and tumor samples with LIN28B reactivation.

Related to Figures S3. (A) Boxplot showing the level of 3ʹ polyuridylation for the two subclasses of let-7 precursors from DIS3L2 CLIP in HEK293 cells. Wilcox rank sum test was used to evaluate the difference between the two subclasses. (B) Quantification of 3ʹ polyuridylation of let-7 pre-miRNAs from LIN28B eCLIP in K562 cells. The y-axis shows the total number of unique uridylated CLIP tags expressed in reads per million (RPM) that overlap with each pre-let-7. The x-axis shows the number of mock CLIP tags (input) expressed in RPM reflecting the abundance of the pre-let-7. The difference between LIN28-mediated uridylation after controlling for pre-miRNA abundance is tested using ANOVA. (C) Changes in the expression of mature let-7 miRNA upon perturbation of LIN28B levels (overexpression or knockdown) in HEK293 cells. The boxplots indicate the interquartile range of each subclass. The difference between the two subclasses was evaluated by a t-test. (D) CSD⁺ let-7 miRNAs showed stronger downregulation by LIN28B than CSD⁻ let-7 miRNAs in multiple types of tumor samples. For each tumor type, average distance correlation (dCor) estimated between LIN28B and miRNAs from each subclass are given on the left (hollow bars); the sign is designated by Spearman’s correlation, and p-values estimated by Mann-Whitney U test. Error bars represent SEM. Pooled reads across miRNA classes produced total expression per class and their dCor with LIN28B expression is given on the right (solid bars) for each tumor type; the p<0.01 cutoff, estimated by permutation testing, is given in broken gray lines. (E) The response of CSD⁺ and CSD⁻ let-7 miRNAs to changes in LIN28B expression in tumor samples. Samples are binned into 20 same-size bins according to LIN28B expression. Each bin is represented by the average fold change of total expression in each subclass relative to the first bin across samples in the bin, and curves were fit to a polynomial distribution with order 3. Similarly, LIN28B average expression fold changes are given on the right axis. Error bars represent SEM.

Figure 4:. The proposed model of selective let-7 microRNA suppression modulated by the bipartite LIN28 binding.

Related to Figure S4. CSD⁺ let-7 miRNA precursors have both CSD and ZKD binding elements, which efficiently recruit LIN28, leading to their 3ʹ uridylation by TUTase and degradation by DIS3L2 (arrows with solid line). CSD⁻ let-7 miRNA precursors lack (U)GAU binding element and are recognized by LIN28 with lower binding affinity, leading to less efficient or partial suppression of these miRNAs by the LIN28/TUT/DIS3L2 pathway (arrows with dotted line), allowing them to enter the DICER processing and RISC incorporation.